Application of Reinforcement Learning for Intelligent Support Decision System: A Paradigm Towards Safety and Explainability

Artificial Intelligence (AI) offers the potential to transform our lives in radical ways. In particular, when AI is combined with the rapid development of mobile communication and advanced sensors, this allows autonomous driving (AD) to make a great progress. In fact, Autonomous Vehicles (AVs) can mitigate some shortcomings of manual driving, but at the same time the underlying technology is not yet mature enough to be widely applied in all scenarios and for all types of vehicles. In this context, the traditional SAE-levels of automation (J3016B: Taxonomy and Definitions for Terms Related to Driving Automation Systems for On-RoadMotor Vehicles-SAE International. Available online: https://www.sae. org/standards/content/j3016_201806/) can lead to uncertain and ambiguous situations, so yielding to a great risk in the control of the vehicle. In this context, the human drivers should be supported to take the right decision, especially on those edge-cases where automation can fail. A decision-making system is well designed if it can augment human cognition and emphasize human judgement and intuition. It is worth to noting here that such systems should not be considered as teammates or collaborators, because humans are responsible for the final decision and actions, but the technology can assist them, reducing workload, raising performances and ensuring safety. The main objective of this paper is to present an intelligent decision support system (IDSS), in order to provide the optimal decision, about which is the best action to perform, by using an explainable and safe paradigm, based on AI techniques.